“Why are we doing all this?” - the creator of Prisma and the former leader of VK projects about his new secret project



    Remember the Prisma app? In 2016, it seemed that every second photo in the world was missed through it. The history of the rise and fall of its popularity was also discussed everywhere (including on Habré ).

    But in June of this year, the founder of Prism, Alexei Moiseenkov ( darkolorin ), together with the co-founder, left the company without commenting on the reasons for leaving. Almost immediately they became known and so. Alexey has launched a new startup and has already raised $ 1 million in investments for it.

    The company is called Capture Technologies inc., But it doesn’t say what it will release. It is known that this is a kind of “analogue of a social network in which the camera and artificial intelligence technologies will play a big role.”

    We are with fillpackart they asked for a conversation and tried to find out, at least that this project was under the hood, but they again fell into the “to be or not to be” discussion.

    About Alexei Moiseenkov



    Aleksey has been programming since he was 11, at school he was a professional player in World of Warcraft (he still plays in Hearthstone), he somehow managed to earn money by teaching roller skating, then he studied at St. Petersburg Polytechnic at the faculty of technical cybernetics. He worked in Yandex, Mail.ru, interviewed by Google. He taught at the Moscow Institute of Physics and Technology, but said that the students did not trust - “I didn’t have my own startup”.

    Later Prisma was launched and for several months was in the top of the AppStore almost all over the world. Millions of users, the constant attention of the media, travel to Silicon Valley. It sounds nice, but Alex seems to be looking at it more critically:

    “Any interview is a distortion. No one is interested in the truth, any media adapts to the reader and fills the story with fiction. The story that, for example, two dudes were sitting and doing a startup at computers and all they needed was computers that would never be popular. Their program may be popular, but the media will always come up with something about its creation. ”

    But we are not like that! We just need computers.

    Are you looking at the code now?

    Aleksey: I wouldn’t say so. Only for myself some things on Deep Learning I watch how to train the network. I understand mathematics well, because I like mathematics, but I do not write anything in production.

    But it helps to work with the team?

    I can read the code, I understand the projection patterns. You technically understand what is being done, you can delve into it. But still, it’s more about a common language with the team.
    In general, I have always performed a semi-technical role wherever I worked - in Yandex and in Mail.ru. I always understood what was happening around.

    And who did you work there?

    In Yandex, I worked on the last courses of the university; I was a manager in Maps. At Mail.ru, I worked at the junction of product analytics for several projects. After I switched and began to deal more with products - vision, strategy, and so on. And after that was Prisma and a fracture occurred.

    "Prism" you do not combine with work?

    At the beginning it was such a project on the side - even more likely an idea for a side project. Fultaym began only later, when the application appeared.

    Phil: You started Prism when there were no neural networks around at all, right?

    I would say that there was simply no such hype. It is now everywhere and masks, and tracking, and segmentation, and other things. And then it was not so popular, probably because it required a lot of resources. Even the style-transfer task was carried out for several minutes. No one will wait for five minutes one picture. But I had an understanding that neural networks would definitely go uphill, so I began to work on them.

    About Oleg Illarionov



    In the spring of 2018, Oleg Illarionov took the lead from VK. He has worked there since 2010 and has probably seen all the most important stages in the life of a company.

    “Pavel is sitting through one office away from me, just on the way to a table with food, near which we all often discuss solutions to various technical problems,” he said six years ago.
    In addition to the main features of the social network, Oleg was involved in launching the Vinci services (clone Prisma), Snapster and Stickerface.

    Like Alex, he did not comment on his departure at first. And only recently announced that he became the technical director of Capture Technologies.

    Oleg, did you have a feeling that you can no longer write the grocery code?

    A: Actually, no, I'm still writing code. It was just a desire to engage in applications fully. That is, I did not want to give up programming - I wanted to take on even more responsibilities.

    F: Didn’t you have a feeling that because of other responsibilities you cannot already do programming well?

    A: Of course, I can’t do it as well as if I had just written the code. On the other hand, in VK, we have always devoted a lot of time to the product. When there was a task to make any feature, redesign, friends section or something like that, I had to communicate with the designer myself, decide how it would work. In general, I had a grocery background almost from the very beginning, and only wanted more of it.

    I read your old interview, of the year 2011. You already worked at VK and told a lot about it. So, you found all the most important milestones in the company. Tell me about it?

    A: It is true. I managed to see, I would say, very different companies during this time. I settled in one, went a few months ago completely out of the other - in spirit, in character, and in everything.

    VK smoothly changed over the years, there were all sorts of transition periods when no one understood what the company would be in a few years. They understood that she would not remain a small startup, but they did not even know what would grow out of her. I would mark three stages.

    The stage of rapid development was when I came there - in 2010. It lasted until 2012–13. Then we just experimented, sat in the office from morning till evening, did all sorts of features. They did a lot of things.

    Then there was the stage of rejection of features. We focused on key things, looked at the metrics and tried not to get dispersed. This is probably from 2013 to 2014.

    And there was a stage of great growth - from 2014 to the present moment. VK turns into a big company, with a large number of people, with a complex corporate culture.

    It turns out, the growth began when Durov left?

    A: Yes, about then. The company began to grow and hire new people. Prior to this, the size of the state was almost unchanged. There were not very many people, and when someone said that he did not have time, could not cope - the option of hiring a person to help was not even discussed. There was a start-up approach with limited resources.

    These are two different approaches to development - you can make great cool products with a small team, and you can not limit yourself to resources and hire a large team. Each of the approaches has its pros and cons. With a large team you can cover a lot of things, with a small one you can make a very polished product.

    A small team is easier to manage, easier to pay attention to the important. In large companies, many things slip out of sight, fall into the seam of responsibility, when it is not clear who is responsible for these things. As a result, there are cases when the key things for business just fall out of sight and do not develop.

    F: I worked in one of the giants of the industry as an ordinary developer and spent the lion's share of time searching through mail and all kinds of communications. I'm afraid to imagine how much time a project spends on such things.

    A: With the growing number of people in VK, more and more time was spent on this. When VK was in a startup format, we used a very interesting approach to solving problems - once a week, on Fridays, for about an hour the whole team gathered in a room (by the way, in a small one, and everything was placed there). We had a projector and a screen, everyone came out and just told, showed what he had done. Option not to go out and tell almost was not. It was necessary to explain how the week went, so everyone worked very productively.

    In a large company, I can easily imagine people who do not work at all for a relatively long time, and no one even notices. In a small company it’s not something that doesn’t work - it’s impossible to work even relaxed.

    Burnout and success




    There are things that you think about - become my startup a huge company, such a garbage will never be there.

    A: I came to Yandex in 2013, and during my time I hired about one and a half thousand people. There was an internal schedule, and it was possible to see in real time how people grow. Even in our office people were added.

    When you start to hire hundreds and thousands, make it so that you control everything, everyone comes and tells something - it is simply impossible. I am deeply culturally convinced that the problems at this stage cannot be avoided.

    Another question is whether management can build a culture (how the values ​​are heard from top to bottom) so that people have the desire and ability to perform tasks when they are not ashamed to talk about their problems and difficulties to people sitting next to them. It’s not something that comes up and reports that today closed three tasks - but that there is openness. To immediately say from the threshold: "we practice transparency."

    If this transparency is brought up by the people hired - even if not by everyone, because it is impossible for everyone - then this will greatly simplify growth. People will get used to do, get used to talk about their problems, about their expectations. For me, this is important.

    But I do not believe that now I will make my blackjack and everything will be as I want and everything will be perfect. Growth problems are natural and impossible to avoid.

    By the way, you told me that before “Prism” it was very hard to burn out at work.

    A: In Yandex, I was engaged in universal management. I did not have a specific list of duties, so I did everything. And about a year later there was such a personal feeling that I burned out of it.

    What thoughts then most tormented?

    A: Probably, I was tormented by the fact that I was standing still and did not understand whether I was ready to go further. I could not objectively evaluate myself. It seemed that nothing was happening. There was always a very big question in my head - why do we all do this globally? It is very difficult sometimes to get an answer to this question. It tormented me a lot in my time. I came to rallies and did not understand what was next.

    It was a combination of the fact that I do everything, the same thing, and I do not understand where all this is going. I ate myself in my head all the time. It is rather personal when there is no answer to your own questions. You can not answer them and start to burn.

    And when Prisma happened, and it became clear that something cool was starting - was it cured?

    A: The feeling that something cool is starting has appeared only when a wild growth has gone. And when we released the application, it seemed that no one needed it. For about a week there was absolute, mortal silence. It could be a hundred or two hundred downloads.

    This is the transitional moment when you are sitting and this is a pancake, is it really not necessary for anyone again? So what to do with it?

    And then at some point, rapid growth began, and there was an awareness that a lot of things should be done right now. A lot of things. And it gave a drive. When millions of people use your product, and you have a small team of five and a half people, everything grows, people around are trying to find out something, ask. And you still have product challenges - this is the drive, and it becomes interesting. Such an experience is difficult to get somewhere else.

    We just recently spoke with guys who returned from the valley after the accelerator and received investments - they are full of enthusiasm. What was your feeling after the valley?

    A: I had a little bit different. The eyes did not burn because we were in the valley, and we recognize people, but because something unthinkable is happening. I had three million downloads per day. I could not imagine such a situation. In Russia, we had 350 thousand downloads per day, and the top AppStore at that time was 13 thousand. Do you understand what the gap? And so for several months. We just sat organically upstairs without anything, no marketing effort.

    Catching such a wave is, of course, priceless.

    How were Vinci and Prisma




    I think you constantly did something similar - Prisma and Vinci, Sticky AI and Stickerface. Honestly, who was looking at someone?

    A: To be honest, we never hid what we saw. Vinci began after the release of Prisma. We saw and decided - need to copy. As for the Stickerface, these are generally different things, and nobody has ever looked at it. Moreover, the idea of ​​Stickerface came to mind a very, very long time ago, when no one did anything like that. If you could manage to take it with all your hands and head, then maybe something would have happened. And now there are a lot of players.

    How did you come together in one company?

    A: We started talking when Vinci was launched. To my surprise, Alexey responded very positively to this. I thought he would be mad, and he somehow perceived with a smile that we copied the application. This is where the communication originated.

    A: I still realized at university that I was not afraid of copying. I think this is normal. Therefore, I had no heite and hatred.

    Who do you think, technically, it’s better to do this thing?

    A: I think it's cool to do something first, but copying is not cool. I don't like to copy at all, and I like to do original things that no one has ever done before.

    In spite of the fact that during all my work I was engaged in some non-original projects, from a technological point of view I always tried to work on the problems that had not been solved so far.

    A lot of things from VK had to do when there was no information about it at all. I had to invent and invent. In Snapster, despite the fact that the idea itself is secondary, there was a full-fledged adult color correction with deep immersion, which no one before us did in the mobile application. We released a relatively unique thing at that time. Another issue is that technical implementation was not enough to disrupt the market.

    In Vinci, both the idea and the technology itself were secondary. But it was an interesting experience, because nowhere was it described how to do it. It was necessary to understand and face unobvious tasks.
    Oleg's report on the technologies used in Vinci before the start of offline processing on the CPU.


    The first version was made in 24 hours, release - in 2 weeks.

    Stack: Torch to work with neural networks, but with a lot of problems. Lua / Turbo for communication between the front-end and backend - and with them, too, everything is not smooth. Backend was first on NodeJS due to lack of time, then rewritten to Go.

    Iron: at first they considered the option of virtual servers with video cards, but decided to buy their equipment. Choose between four options from Nvidia - Tesla M40, Tesla K80, GTX 1080 and Titan X. Chose the latter two, because the main advantage of industrial options (Tesla) in fault tolerance, and this problem has already been solved by architecture.

    Architecture:Each style is processed on its video card. The distribution of styles on video cards is dynamic and is regulated by a special function.

    The most unexpected botlnek - the video card could process more images than the network missed.

    Organization of training: put FGlab on all servers, began to bombard each picture with random parameters, and manually choose which ones you like more. This helped make the styles very different from each other.

    The ratio of the speeds of the layers: input - 42%, intermediate - 10%, and output 28%.

    And how was Prisma arranged under the hood?

    A: The models were on the "Torch". This is an old thing that is not very popular now, because there is PyTorch and TensorFlow. The model itself was packed in video card processing. She chased away Inference, and from there “Python” pulled everything upstairs. That is, the backend, the wrapper on Python, the queues and the server part are also on Python. There was no hard stack. On the device, just a thin client, interface plus work with api.

    We went through two stages. The first is when everything was online. We initially accelerated the algorithm about a hundred times from the backpropagation version, running it on rented gaming video cards. The difficulty was that with the growth of the application there were queues. The flow of image requests was huge. Pictures went to the server, and with a load of several million downloads, it was very expensive to drive terabytes of photos. Paying for the server and for the traffic would have to be more than for the resources - even if we reduced the waiting time.

    So began the second stage. We thought, is it possible to make offline or on some cheap processor processing. We managed to do this in August 2016, and by the end of September we began to deploy. Then there was no Core ML, no processing on the phone. We were the pioneers of Metal, built all of the crutches and sticks, but we understood that this was the future. Because it saves privacy - does not drive photos of the user anywhere, and this grandly reduced the cost of processing.

    The transition was key for us, we were able to implement a family of algorithms for processing models on the device. And we began to understand that the entire processing industry, AI, Machine learning - everything will be on the device, with the data that is on this device.

    F: So, at the beginning of work on Prism, the idea of ​​doing calculations on the client was not even considered?

    A: Not exactly. We looked in that direction, but thought like “Oh, yes well! This is unreal. In the beginning, even these models hardly fit into the memory of the GPU, and there are 8 or 12 GB. The idea that this was possible on the device seemed futuristic.

    But we have gone up - everything is on fire, the servers are on fire, there are millions of users. The focus has shifted to the system that we built on the GPU. We have optimized the server part for a month, but realized that without processing on the device, nowhere. Otherwise, you will go bankrupt instantly.

    I did not think that there would be such a flow of images.

    What load did you expect?

    A: Before launch, we thought there would be a maximum of three images per second. Maximum! From here you estimate - 60 x 60, 3 x 600, for another 12, and it already seems - a lot, where so much! Imagine how many users should be and think, three images per second - this is normal.

    We ran the processing, and the whole cycle went 400 milliseconds. Thought it was great. As it later became clear, this is not at all great. When millions of users poured, all thoughts began to go completely the other way.

    And what did you do? Ran, strangled?

    A: When it was burning, we used all possible resources on the market. All they could find. I remember, we even called friends, asked if they had servers on video cards. I called almost everyone I knew. Developers from “Yandex”, still from somewhere, said “damn guys, help someone to raise a pair of wheelbarrows”. They took 5-6 machines from someone, and they were not even on gaming video cards, their processors were slower.

    We called Amazon, where we were given an increase in limits. Seriously, when we had four cars of four video cards on Amazon, we were so happy - a line of ten images fell to eight. We were partners with Servers.com - they also really helped at that moment.

    F: But was it that the code was written poorly, and you had to correct and rewrite after the release?

    A: On the fly there were a lot of fixes, seriously. Fixed precisely the bottlenecks that broke, where it was impossible to expect such a load. After all, everything was written to just work. At first there were no lines at all. It is probably difficult for me to evaluate the purity and literacy of the code. Since everything was written literally in a month - yes, probably there was no industrial approach, as everyone likes, taking into account the loads, taking into account the extensibility. Therefore, a lot of things corresponded.

    What will happen to AI, ML and neurons


    F: When I figured out how neural networks work, I lost the feeling that this is magic and a breakthrough. And you?

    A: Actually, yes. For me, all deep learning is no magic. Differential geometry is magic, and deep learning is quite transparent and technically understandable. From the point of view of which algorithms are inside and how they work, there is a certain black box built on the heuristics of experimentation. But in general, the thing itself at that time was definitely not something supernatural. It was necessary to believe that this would work quickly on devices, and when we achieved this, I only consolidated my convictions.

    So with any technology. Let's take databases in the nineties. It was also a feeling that nothing complicated, but somehow it was going hard. And when everyone understood that it works, it is useful and generally the only thing on which data should be stored, the database has become an absolutely normal thing, now everyone uses it. The same thing should happen with ML-models.

    F: Suppose I saw a rather abstract AI in the garage using neural network algorithms. And now the feeling does not leave me that if I study properly how this is done, then knowledge will narrow the circle of possibilities that I will decide on. Because I stop believing in the possibility of technology. You do not have this?

    A: There is no common AI for everything. Even combinations of technologies are rarely represented. A good example of several technologies in the bundle is Google Duplex. He listens to what you say, and tries to deduce the context, and tries to analyze this context somehow, plus generates speech. This is a whole complex, but even it remains a very narrow history - it calls and makes reservations. That is, it performs tasks with clear scenarios and sets of accurate, well-described tasks.

    It seems to me that now the stage of development of AI and ML is when they are able to perform only well-defined, well-studied tasks for which there is a sufficient set of data. But I have not yet seen a more general application.

    A sense of the possibilities of technology runs ahead of its real possibilities.

    A: Yes. But I think soon we will see attempts to make a combination of algorithms and systems that will allow to perform more abstract tasks.

    F: I often heard that we stole the image processing implementation from the biological mechanism. Do you think so?

    A: No, actually it is not. When there was a boom of neural networks that flared up precisely because of convolutional neurons, it seemed that this was similar to human vision, and that everything would develop in that way. There are certainly similar patterns there, but to say that all one-to-one works equally well, definitely not. From the point of view of neuro-technology, convolutional networks are one small piece. And there are still a lot of different types of networks.

    F: Do you think such an approach has the right to exist — to try to understand the natural human mechanisms and copy the implementation from them?

    A: Based on my observations of the latest trends, now, on the contrary, technologies in deep learning are moving away from this. There are all sorts of tricky solutions, not like biology. But from time to time there are attempts, articles, and so on. I do not know what we will come to in the end.

    Deep learning has certain tasks - to improve the quality and improve the result. And in what way it will turn out, as with a person or not as a person, it does not matter anymore. This is such a freely developing area, where, as it will, it will turn out.

    And now everyone is discussing capsular neural networks. Have you tried to do something with them?

    A: So far, unfortunately, this is not suitable for some use that is close to production. There are still no adequate and fast implementations. Those examples that exist work with very small simple datasets. I think the community has to do a lot more work with them to make it happen. But the idea seems to me very correct, as in the whole direction of development.

    After all, you have also been working with firestorms. Is it true that other algorithms win there?

    A: Well, not exactly to win. For example, MSQRD is built on decisive trees with various tricks. But now, probably, everything goes to neurons. They have become as fast as previous implementations, but the potential to increase the quality there is more. I am not so deeply familiar with tracking, but it seems there is some difficult solvable bottleneck there. And in neurons, due to the fact that they have learned better how to use them, the power of the devices has increased, the phones have become more powerful - the quality is growing release after release. Therefore, it seems that they have much more prospects than the previous approaches.

    What is under the hood of a secret application



    What technology are you relying on now?

    A: Now we decided to move away from the standard approaches to building the backend and build everything on FondationDB, which recently appeared in the open source. We will have more backing than Prism or Vinci. We are doing a realtime application, and this will require work on complex optimizations.

    The main thing here, of course, is not FoundationDB itself, but its approach — the transactional key value of the repository, in terms of real transactions that are synchronized between servers. This approach seems to be some kind of future for backends. We chose FoundationDB as the most promising storage with such characteristics, we will test on it, and then, perhaps, we will move to another storage, if something will show the results better.

    This is with regard to data storage. Next we have everything on Go now. Vinci was also written on Go. Lyosha said that Python was originally in Prizme, but then they also rewrote him on Go. It seems to be a good language to do backend now. I see positive reactions from the community. It is much easier to hire people, and in general it seems that this is the right decision.

    From an infrastructure point of view, we’ll try to build everything on Amazon. I think in this respect it makes no sense to save and look for something else. We must first start, to understand whether to save. But in general, the Amazon is not so high prices. They periodically index them.

    Accordingly, due to the fact that we run on a shared hosting, the architecture will be different from such projects and applications as VK. In VK, many tricks emanated from the full life on the server. There, on the same server, very different roles could be performed - they could both store pictures, contain a database, and do something else.

    We all break into instances, all very atomically. I set myself a goal to make a non-falling backend. It would be desirable that at any fall of iron, up to a data-center in Amazon, users did not notice anything.

    I will not promise to succeed. But I really want to embody the architecture that is most protected from falls. I'm not sure that I can tell you more without burning our concept.

    We are not afraid that someone will copy it. Just want to endure some kind of intrigue.

    F: Could you tell us more about the motivation to use Go? In addition, it is more convenient to hire developers?

    A: In general, language is not such an important thing. We are not the ones who will do only in such a language and no longer in any. If someone wanted to do in PHP, we would do it in PHP. But right now, it seems that Go is the best option on the market, the coolest thing.

    F: How is it better for a startup to choose technologies? So you decided to write on the “Noda”, but you have 5 million people and that's all - your Noda does not stand up, and now it is very expensive to rewrite. How to predict this?

    Nod did not want to choose a little for another reason. I would not say that Noda is very slow. If the code is well written, it will work well.

    F: There's less space for optimizations.

    A: I would be afraid to do back-up on the node, because it's very easy to make a mistake on it. When something urgently needs to be done or corrected, unpleasant situations can arise. And even if you use TypeScript, everything is clever, typing and so on - all the same, difficult situations occur on the node, bugs that are difficult to catch.

    And Go is a very straightforward language. He does exactly what you write. He does not give space for all sorts of abstractions and tricks. With it it is much quieter to write a backend. You can at any time take and remake a large piece of code, and everything will be fine. If something is wrong, it probably will not compile, and not so that everything will not work.

    F: So you think that static and strong typing is the determining factor?

    A: It is very important for a startup, more important than performance. It is important that you can quickly fix bugs without creating new ones. To minimize risks.

    If you use modern patterns of development on Noda, especially if there is some third-party library, then it is very easy to contact a memory leak. And then find her whole story. In the meantime, you are looking for this leak, you waste time, your backend drops ten times a day. I just want to avoid such things, this is a very big danger for a startup.

    There were many applications that were launched, and the team, instead of sawing new features, supporting and supporting interest in the project, was engaged in fixing a million bugs of all sorts. The application was unstable, and it killed a cool project.

    What kind of people are you currently recruiting?

    A: On the back end, we have not yet begun to recruit anyone. I wrote the whole backend myself. And so, I have a little google approach to hiring people - it’s not very important what a person knows right now. The main mind, learning and good knowledge of their language. In times of rapid growth VK I interviewed a lot of people, and always looked at it.

    The new employee will still have to spend a lot of time to understand the backend architecture. Any language can be learned - it is not so difficult. Therefore, we would come up with any backend developer who simply understands how databases work, which is smart and smart enough.

    From the point of view of neuron other conditions. We hire people already with certain skills. In our case, to teach someone and wait will not work. In neurons, the training period is very long.

    With mobile developers, attention to detail is important. The developer who makes the interface should himself notice and correct many little things. Otherwise, you have to poke him all the time, and it will eat a lot of time and nerves.

    In VK you were not limited in resources. I imagine it as: it’s necessary to see Tesla for $ 5000 - at least ten. It is necessary server for 700 dollars a month - we take. Can you afford such experiments now?

    A: It doesn't look like we had resource problems. About VK is true. When we realized that we needed servers with vidyashki, we were told - no question. They rented almost a private plane to deliver these vidashki as quickly as possible. In this regard, the hands were always untied, and no one spared iron.

    But on the other hand, we now want to use Amazon, not because there are no resources.

    But with a private plane you are no longer lucky with the server.

    Well yes. But on the other hand, there is no need.

    What is the point in startups




    You want to keep the intrigue around the project, and I promised not to meddle with questions. On this occasion, I remembered one story. I read on reddit how a guy got on a closed presentation of a VR headset, and couldn't talk about it because of the embargo. Therefore, he wrote a review of his dinner. But with a trick, like, “the plate with food was a bit heavier than the HTC Vive, but in my hand it was quite comfortable.” Can you tell me about your dinner last night?

    Well, we do not do iron, we do software, you can’t compare it with dinner.

    Damn, the attempt failed. But the fact that it will be a social network - you have already said.

    A: Probably, it is wrong to call it a social network. Rather, it is a product based on some kind of social interaction.

    Was it difficult to convince people of this idea?

    A: It is important for investors to tell not just an idea, it is important to tell who is in your team.

    In general, the likelihood that you will be given an investment is made up of who you are in a team, what is your idea and whether you are going to the right people.

    Going to random people and just taking money from them is probably not very correct. They may misunderstand what you are doing. Raise money on the stall from people who build rockets? They do not understand the business of selling shawarma.

    Therefore, it is important to follow the investment to the right people who are relevant to the area in which you will work. I did just that.

    If you touch the team, for me it is very important that a person likes what he does. Difficult to describe the idea or not difficult, depends on how the person himself understands this area. Sometimes opinions converge, belief in something converges, and you work with such people.

    The success of social platforms very often depends on their image, stereotypes, audience, which is formed around them - even, perhaps, more than on technology. Doesn't that scare you?

    A: This is such a combo factor, I agree. There is a cultural moment, a combination of how you present it, how you tell the market. Why it is needed, how it makes life better. Many say that they solve problems with their products. I tend to do better what has already been done, to simplify.

    There is a combination of filing and execution. And then comes the hypothesis testing - is your assumption correct? Any product is based on some assumption. Take Dropbox: the hypothesis is that people need to store files in the cloud, there is no space on the computer, and you need to have access everywhere. They did it and checked it out. It turned out that someone needs it, people started storing files in the cloud. The hypothesis was confirmed.

    Our hypothesis, too, is that the world needs it. I do not think that any social factor is important here, because it is already embedded in the hypothesis.

    Every successful social thing is also a big responsibility, even the impact on the world. Recall only the recent history of how Zuckerberg reports to the senate. Are you ready for such a responsibility if everything shoots?

    A: Hard to say. Must first shoot. The hypothesis must be confirmed and survive the growth stage. Zuckerberg, obviously, was not responsible to the Senate in the first year of his work, and not even in the third, and not in the fifth. Probably by the time we learn, if such an opportunity presents itself. I would like to introduce myself.

    Oleg, I’ll go back to your report on Vinci. If you change its structure, you can achieve some comic effect. For half an hour you talk about advanced technologies, engineering solutions, architectures, expensive hardware - and then at the end: “that's what we all did”. And on the photo just appear patterns.

    Hence the question - do not you think that the best minds of our time are engaged in trifles?


    (Then there were very long seven seconds of silence. It turned out that Oleg lost the connection, so Alexey entered)

    A: I think this is not quite so. The IT industry has appeared not so long ago. It was formed only by the end of the eighties. The Internet appeared, as a kind of infrastructure project, which grew out of military technology. On the basis of this, a new area was born, and at the moment we see that it is penetrating into all spheres of life. This gives rise to a large number of opportunities, but the entry threshold increases.

    To start a simple thing in your opinion, with a simple result - a picture with a pattern - you need to know the whole stack of technologies, how it all works, how marketing and distribution work.

    And to assert that the best minds are fighting for some kind of fake value is probably not entirely correct. Just the entry threshold requires more knowledge and skills. It is difficult to understand in advance where the value will actually be born.

    Facebook in the beginning was not a brilliant technical idea. They just made up a saytik, did some basic things, and it went away. And then there are technical difficulties and millions of users.

    I think it seems so because the industry has grown, and it has become harder to do simple things than before.

    (Here Oleg appeared and agreed with Alexey).

    What are you doing startups for?

    A: I just like to make such applications. Everything develops, and we get a big and good effect for the industry as a whole.

    And why are we not all involved in launching space rockets, but making some social applications? Everyone should do what he likes. Alexey and I like social applications, and we should deal with them. If someone likes to launch rockets, then as far as possible you need to do this.

    Just do not underestimate the social applications. In fact, this is a very important thing, it has a huge impact on society.

    A: I agree with Oleg, it is important to do what you like. It sounds like a cliche, but I want to make life a little better, I want to help do some routine things faster, better, so that people will have new opportunities. For example, Instagram gives you the opportunity to post a photo, but before that it was not. You can see for yourself how Instagram affects our life. Previously, it seemed an uncomplicated thing, and that's what it has become now.

    A: Instagram is probably not the best example. A good example would be Facebook. Not in the sense that we are now doing another Facebook. I want to think that we are doing a fairly useful application that definitely carries more social benefits than Instagram.

    If you constantly make the world better and better, will it not lead to the opposite effect?

    A: No one knows. If I knew that it would lead to bad things, it might not. And maybe he did. No one knows the future.

    Also popular now: